JPS63265377A - Production of dictionary for optical character reader - Google Patents

Abstract

PURPOSE:To easily produce not only a dictionary for printing types but a dictionary for peculiar handwritten characters with use of an optical character reader itself, by writing laterally the characters to be registered on a single line of an original with addition of a mark showing the height of the line. CONSTITUTION:When a dictionary for peculiar handwritten characters is produced, a character, e.g., 'i' to be registered on a single line of an original is written laterally up to about 10 pieces at intervals and a mark M is put at a position slightly distant from the last 'i' to show the height of the line. Then this original is scanned by an image scanner 2 and the lines are segmented with the space during which the mark M is detected and not in the next time defined as the height of a single character. As a result, even such characters like 'i', 'j', etc., which are vertically separated into two parts can be defined as a single character as a whole. Thus a character pattern can be correctly segmented. In the character register mode even the peculiar handwritten characters can be smoothed and registered in a character recognizing dictionary file of a hard disk device 7 via a keyboard 1. In such a way, a dictionary for peculiar handwritten characters is easily produced.

Description

DETAILED DESCRIPTION OF THE INVENTION Technique 13υd This invention relates to a dictionary creation method for character recognition in an optical character reading device, generally abbreviated as rOcRJ.

Moxibustion 4 Scan a document with image information including characters printed or handwritten on paper using an image scanner, import the image information of the document as image data, recognize the characters from the image data, and convert it into character code data. Various types of optical character reading devices have been developed.

This optical character reading device can be used as a means of inputting character information into word processors, automatic translation devices, processing systems that handle characters such as 11yX aggregation devices and search jp data file creation devices, and communication systems such as data communications that transmit character data. If used, it is possible to significantly improve input efficiency compared to keyboard input.

This optical character reading device is provided with a character recognition dictionary in which the image data of the character phono 1- is registered in advance as reference image information, and the character recognition means refers to the character recognition dictionary and inputs the image data. By comparing the image data of the selected character with the image data of the dictionary and performing pattern matching, this is recognized as a specific character and the corresponding sentence and character code data are generated.

There are various types of character designs, ie, fonts, for commonly used typefaces.

Therefore, it is necessary to provide a character recognition dictionary for each set of a plurality of commonly used character types.

However, in the past, in order to create such a dictionary for character recognition and maintain it by modifying or changing it, it was necessary to use a separate dictionary creation and maintenance machine, so anyone could do it. New dictionaries could not be easily created or modified.

In addition, when creating a dictionary for one handwritten character, there are many constraints, and it is not possible to directly register each individual's quirky characters in the dictionary.

This invention solves the problems in the conventional character recognition dictionary creation method, and uses the optical character reading device itself to create not only printed dictionaries but also handwritten dictionaries. The purpose is to make it easy to create.

2-Hiro In order to achieve the above object, the present invention uses an optical character reading device as described above, in which a large number of the same characters are listed in one line and the height of the characters in that line is adjusted as shown in FIG. A document (A) with a mark M shown in the figure is scanned by a scanner and its image data is read (B), and the dot patterns of each character within the height regulated by the mark M are superimposed or averaged to obtain character pattern data. Perform processing (C) to create one character pattern, and display that character pattern or any character in the read line (
D), give a character code corresponding to the character pattern (E), and register it in a dictionary file for character recognition (F).

Hereinafter, a detailed explanation will be given based on one embodiment of the present invention.

FIG. 2 is an external perspective view showing an example of a document processing system that can be used for a word processor, an office computer, an automatic translation device, a form processing device, etc., which has the function of an optical character reading device embodying the present invention.

This document processing system has alphanumeric keys, character keys such as kana keys, cursor movement keys, various function keys, etc. as input devices, and includes a keyboard 1 for inputting instructions from an operator, and optical scanning of JM manuscripts. The image scanner 2 inputs image information including characters as image data.

In addition, as an output device, a CRT display device (hereinafter simply referred to as rCRTJ) is a display device for displaying various text and image information including guidance for the operator.
3, and a printer 4 such as a laser printer for printing out various information processed by this system.

The main body 5 includes a floppy disk device (FDD) 6 and a hard disk device I, which are data storage devices.
(HDD) 7, and as shown in FIG. 3, a control unit (CPtJ) 10 consisting of a microcomputer etc. that centrally controls the operation of the entire system, and a program memory R○M 11. R which is data memory
AM (256Kbit.

Above) 12. Keyboard interface 13. Scanner interface 14. cItT controller 15, FD
D controller 16. HDD controller 17. A printer controller 18 and the like are provided.

With this system, the image data of the original image read by the image scanner 2 is transferred to the scanner interface 1.
CRT! The data is used to perform processes such as typeface discrimination and character code discrimination according to the present invention.

Multiple dictionaries for character recognition in which image data of each character is registered as reference image information for each typeface are usually H
It is stored in DD7.

Note that a dictionary can also be created for handwritten kakuji (unformatted typeface) as described later, and it can be registered and used in the same way as a dictionary for printed characters.

Therefore, before explaining the dictionary creation method according to the present invention, the operation of this optical character reading device that uses the dictionary created by the method will be explained with reference to the flowchart of FIG. 4 and other figures.

When recognizing characters by directly importing the image data of a document from the image scanner 2, first, in step 1, the density of the characters written on the document is determined, and in step 2, the image data of the image scanner 2 is determined according to the determination result. Set the brightness of the fluorescent light for document illumination.

The character density can be determined by specifying the density by an operator, or it can be automatically determined from the detection level by partially scanning a document with an image scanner.

The brightness of the fluorescent light is set so that if the text on the document is written darkly, the light is dimmed, and if the text is faintly written, the light is lit brightly.

This prevents one character from being crushed or missing.

Then, in step 3, the image scanner 2 scans the entire surface of the document at a preset reading density, imports the image data into the main body 5, and stores it in the memory (RAM 12).
) 6 On the other hand, when performing character recognition from image data stored in an image data file such as the HDD 7, the data from the image data file is read and written to the memory (RAM 12).

In this case, it is necessary to store the image data read by the image scanner 2 in an image data file in advance.

At this time, the image data file is provided with a header section as shown in FIG. 5, in which information about the reading density and the length and width of the read document is stored.

This vertical and horizontal length is required when cutting out one line and cutting out characters for character recognition.

Reading density information is required when determining character codes.

Next, the process proceeds from step 3 or step 4 to step 5, where automatic line cutting is performed, and then, in step 6, character cutting is performed.

Incidentally, since the image scanner 2 retrieves image data by scanning the original in the horizontal direction, the image data is sequentially input in bytes in the image data file or the image data storage area of the RAM 12 as shown in FIG. ing.

Therefore, when cutting out one line and cutting out characters, it is not possible to cut out the image of one character unless there is information about the vertical and horizontal lengths.

Furthermore, as the reading density increases, the height and width of -characters increase, so naturally the matching data also changes.

Figure 6 (A) and (B) show the same characters at a reading density of 200d.
This shows the state of image data when read at pi and 300dpi.

Also, when cutting out lines, remove horizontal shading,
The space between the lines is cut out as a line, but if the original is set in the image scanner at a slight angle, the read image data will look like the one shown in Figure 7, and even if you take the horizontal shading as a whole, the space between the lines will be cut out. You may run out of space.

Therefore, in such a case, as shown surrounded by thin lines in FIG. 7, lines are cut out by dividing the lines into blocks such that spaces are created between the lines by horizontal oblique shading.

Next, in this embodiment, in order to be able to read handwritten characters, the character type is determined in step 7 of FIG. 4.

In the case of handwritten characters, in order to increase the recognition rate, 1. For example, as shown in Figure 8, the document is divided into multiple fields, and the information on the length of each field and the Character types (alphabet, numbers,
Specify the specified information (symbols, hiragana, kanji, etc.).

Therefore, depending on whether or not such information is present, it is possible to determine whether a character is handwritten or printed.

In the case of printed characters, the process proceeds to step 8, where character font discrimination (printed typeface discrimination) is performed to determine a character font dictionary to be used for character recognition.

In step 12, character code determination processing is performed using the dictionary, the details of which will be described later.

In the case of handwritten characters, the process proceeds to step 9 where smoothing processing is performed to correct irregularities, and in step 10 the size of the characters is corrected through normalization.

In this case, for example, the letters are enlarged or reduced to the same size within a range that does not make it impossible to identify alphabets in which uppercase and lowercase letters have the same shape.

Then, in step 11, the handwritten font is determined and
A dictionary for handwritten fonts used for character recognition is determined, and in step 12, character code determination processing is performed using the dictionary.

Next, in step 13, it is determined whether character recognition is finished or not.
If it is not the end, the process returns to step 5 and repeats the process from cutting out the first line to character code determination.

This process ends when character recognition up to the last line is completed.

Here, the above-described character font (typeface) discrimination and character code determination processing will be explained in detail with reference to FIG. S and FIG. 10.

The character font discrimination process is performed as shown in the flowchart in Figure S using multiple character font dictionaries. Rank them.

Then, for one line of character image data, matching is performed starting from the dictionary with the highest priority, and matching with dictionaries with lower priority is performed one after another until a matching result of a certain level or higher is obtained.

At that time, if the matching result with a certain dictionary is unsuccessful, that dictionary is given the lowest priority, and the other dictionaries are raised one by one in priority.

- If a matching result of a certain level or higher is obtained as a result of matching with multiple dictionaries in the street, the dictionary at that time is selected, and if that cannot be obtained, the dictionary with the highest matching is selected. Determine character font.

Normally, documents are printed using the same character font, so by prioritizing the dictionaries in this way, it is possible to speed up character recognition (character font discrimination and character code determination) from the next line. can.

To explain this character font discrimination process with reference to FIG. 9, first, the first priority dictionary 1F (A dictionary in the above case) is read, the first character of one line is read, and pattern matching (or feature matching) is performed. If matching is achieved and recognition is possible (OK), register value X
(initially "0"), then subtract a certain value α, and if matching is not achieved, the register value X is subtracted by a certain value β (α minus β
) is added.

Then, it is determined whether or not it is the end of one line, and if it is not the end of one line, the next character is read and the same processing is performed.

When this is done until the last character of one line, the value of X at that time is memorized, and then a judgment is made as to whether or not
If so, the dictionary used at that time (Dictionary A) is determined.

If X<0, it is determined whether or not there is an unused dictionary, and if there is, the priority order of the dictionary is changed, for example, as in the second judgment of the clothes, and a new first judgment is made. Priority Dictionary (B#
``J), and perform matching from the first character of one line to the last character of the line and perform the same process as described above, then determine if X<0, and if YES, the dictionary used at that time ( B dictionary).

If it is not 0 at this time, it is determined whether there is an unused dictionary or not, and if there is, the priority order of the dictionary is changed again, for example, as in the third judgment of the clothes, and a new one is created. Then, the dictionary with the first priority (C dictionary) is read and similar processing is performed.

Then, if X<O, the dictionary used at that time (C dictionary)
However, if it does not become X and O at this time and there are no unused dictionaries, then the
Compare the values of and decide which dictionary is the smallest.

Next, the one-character code determination process is performed according to the flowchart in FIG. 10, and the dictionary determined by character font discrimination is first read, but when this character code determination process is performed following the character font discrimination process in FIG. By the time you reach 0 and decide on a dictionary, that dictionary is already loaded.

You can skip this step.

Then, it reads the first character of one line and performs pattern matching (for example, 24-dimensional matching) with a dictionary. If the matching is successful and the character code can be determined (OK), the character code is output, and the code is determined. If it is not possible (NG), then pattern matching using other methods (4X4X
8-dimensional matching, 3X3X8-dimensional matching, pattern matching using multi-layer direction and stogram method, etc.), and if the character code can be determined, the character code is output, and if the character code cannot be determined, the character code is output. Outputs an invalid code.

This process is performed sequentially until the last character of one line, and 1
Performs code judgment processing for a line.

In this way, character font discrimination and character code determination are successively performed line by line to read the entire document. When displaying or printing the read character data, a mark indicating that there are unreadable characters is displayed or printed in the part where the unreadable code exists.

In this way, the function that automatically identifies the font and reads the surname using multiple character font dictionaries without specifying the font used in the printed text is hereinafter referred to as "multi-font". .

Also includes handwritten typeface recognition and character code determination processing.

Pattern matching is carried out in substantially the same manner as in the case of printed characters, but in this case, pattern matching is also performed using the field length and character type information shown in FIG.

The method for creating the handwritten Kusegyu dictionary used there will be explained later.

Further, it is also possible to perform the character font discrimination and the handwritten typeface discrimination in FIG. 4 together as a typeface discrimination process.

[Regarding dictionary creation and maintenance according to the present invention] Next, a method for creating and maintaining a dictionary used in the above-mentioned optical character reading device will be explained.

In the document processing device having the function of an optical character reading device shown in FIGS.
The utility selection process shown in the flowchart of FIG. 1 is started, and a main menu indicating the type of process is displayed.

Then, when a process is selected manually, it is determined whether it is [Single Dictionary Creation Maintenance Utility], [Multi-font Dictionary Creation Maintenance Utility], [Dictionary File Name List], [Handwritten Dictionary Creation Maintenance Utility] or (End). ”.

Single dictionary creation/maintenance utility> This is a program that creates a single dictionary for printed characters used for character recognition. It has the functions of deleting and printing a list of registered characters, and each reads and writes from the dictionary file.

"File name registration" is executed according to the flowchart shown in FIG. 13, and allocates a file area and registers a file name in a directory.

"Character 9. Record/Add" is executed according to the flowchart shown in FIG. 14 and is the main character creation function.

Here, when you enter the number of characters in one line and the reading density, one work permission is granted, and you set the document, the scanner reads one line of characters and performs one image process to overlap or average the patterns of each character and create one character. A pattern is created and either the character pattern or one line of characters is displayed on the CRT.

When the operator sees the pattern and enters the corresponding character using the character keys.

The character code and the data of the character pattern being displayed are associated and written to a dictionary file.

Note that this character registration will be explained in more detail later.

"Character deletion" is a process that is executed according to the flowchart of FIG. 15 and deletes registered characters in the dictionary file.

``Registered character list printing'' is executed according to the flowchart in Figure 16, and is a process of outputting the characters registered in the dictionary file to the CRT 3 or printer 4 (Figures 2 and 3) for display or printing. be.

Multi-font dictionary creation and maintenance utility> The multi-font function allows character recognition without specifying a font as mentioned above.

This is a file creation/maintenance program that registers dictionary file names to be used.

As shown in Figure 17, this program includes processing programs for registering multi-font file names, registering dictionary file names, deleting dictionary file names, printing registered characters in dictionary files, adding dictionary file names, and changing dictionary file masters. Become.

Each process is executed according to the flowcharts shown in FIGS. 18 to 23, respectively.

In this example, six dictionary file names can be registered in the multi-font file, and the registered dictionary file names are prioritized as described above.

For example, when first registering all new dictionary file names, the priority order is set in the order of registration, and when that is used, the priority order is swapped as described above, and when additional dictionary file names are registered, the priority order of that dictionary is changed. When a dictionary file name is deleted, the priorities of dictionary file names with lower priority than that dictionary are raised and reassigned.

Furthermore, when the dictionary file name is replaced, the priority order is also rearranged in the same way.

List of Dictionary File Names> This is a program that displays a list of file names of character recognition dictionaries on the current disk (the disk on which you are currently working), and is executed according to the flowchart shown in FIG.

In this example, eight dictionary file names can be displayed on one screen (one page), but if the total number of registered dictionary files is eight or more, press the N (Next) key. The dictionary file name of the next page can be displayed by pressing the B (back) key, and the dictionary file name of the previous page can be returned to the display by pressing the B (back) key. and,
Pressing the E (end) key ends this process.

Handwritten Dictionary Creation and Maintenance Utility This is a program that creates a handwritten dictionary used for character recognition.As shown in Figure 25, it can register two characters in a file name to a dictionary file, add and delete two characters, and print a list of registered characters. There are various functions.

This function is the same as that of the single dictionary creation and maintenance utility shown in Figure 12.
The flowcharts shown in FIGS. 6 to 2S are also substantially similar to the processes shown in FIGS. 13 to 16 in the single dictionary creation and maintenance utility.

However, in the character deletion process shown in FIG. 28, a "deleted character type input" is provided, which allows the user to specify the types of characters that can be deleted (3 types: 1 for printed text, 2 for handwritten text, and 2 for both).

Next, the third section explains how to create a printed or handwritten dictionary.
A detailed explanation will be given from Figure 0 onwards.

When desired characters (including symbols, etc.) are registered in a dictionary, an image scanner scans a document in which many of the same characters are listed in one line, captures the image data, and takes the horizontal oblique shadow.

As shown in Figure 31, when looking at each character from the horizontal direction (direction of arrow H) perpendicular to the scanning direction S, you can determine the beginning of the character (the point from the white part to the black part that is the shadow of the character). ) to determine the end of the character (the place where the shadow of the character changes from black to white), determine the height of the character, and cut out lines.

This horizontal diagonal shadow has a certain height in advance, so that, for example, when reading a slightly faded character, even if the horizontal diagonal shadow disappears at the blurred part, it will not be mistaken as the end of the character. Therefore, if the initial height of this horizontal diagonal shadow is set high, characters consisting of vertically separated parts such as li i pg, ``j'', or ``:'' will be recognized as one character. be able to.

However, if you do this, when you try to register a small character, there is a risk that some of the adjacent characters will be judged as one character, so the height can only be set to the minimum necessary range. .

Therefore, in the example described below, the characters (in the illustrated example, a
ver) are written in a string of LO characters, for example, at intervals in the horizontal direction, and a mark M (vertical line in this example) indicating the height of the characters in this line is added at a position slightly away from the last character.

By scanning this document with an image scanner and importing the image data, it is possible to accurately determine the height of one character from when mark M is detected until it is no longer detected, and to cut out lines. So,
Even a character consisting of two vertically separated parts, such as ``91'' or ``pano'', can be treated as one character as a whole, and one character pattern data can be correctly extracted.

Even in the case of a small character, by adding a mark M that matches the height of the character, only the pattern data of that character can be correctly extracted.

In addition, by doing this, when registering handwritten characters, it becomes possible to register even irregular characters, such as writing the dots of "i" too far apart, without any restrictions.

In order to prevent dust or dotted noise from being mistaken for a mark, it is desirable to make the mark M somewhat thick so that black level data for several dots can be obtained during horizontal scanning.

The process for registering a dictionary using this method.

This will be explained using the flowchart shown in FIG.

As shown in Figure 32, write 10 characters to be registered in one line (printed or handwritten), and add a mark M indicating the height of the character a little apart from the last character. When a document is set in the scanner and "reading J" shown in FIG. 14 or 27 is started, the process shown in FIG. 30 starts.

First, start the scanner and start scanning the document.
A predetermined number of dots (mark M
Unless a black level greater than (less than the number of dots detected) is not detected, it is assumed that the white part (space part) of the document is being scanned, and scanning continues without doing anything.

When a black level equal to or greater than a predetermined number of dots is detected during one scan in the horizontal direction, it is determined that a black part of the mark M has been detected, and cutting out of one line of image data is started.
One line is continued to be cut out until the black part is not detected, and when the black part is no longer detected, one line cutout is finished.

Characters are then extracted from the extracted one line of image data, and character pattern (dot pattern) data of each character (in this example, 10 identical characters) included in that one line is extracted.

In the case of handwritten characters, it is desirable to perform smoothing processing to correct irregularities in the character pattern and normalization processing to slightly enlarge or reduce the entire character pattern in order to unify the size.

Next, a superimposition process is performed in which the corresponding dot data (1'' or 0'') of each character pattern are ORed and superimposed. At that time, if the black level data for each dot position is less than a preset number, it is treated as the white level, so that the influence of noise can be removed, and in the case of handwritten characters, the influence of font variations can be eliminated. It is possible to obtain an averaged character pattern by reducing the number of characters.

The character pattern or one line of characters thus obtained is displayed on the screen of the CRT 3, for example, as shown in FIG.

When the creator confirms this display and enters the character corresponding to this character pattern (in this example, "i") using the character key on keyboard 1, a character code indicating that character is generated, and it is input as described above. Add it to the character pattern data obtained in 9. and save it to the character recognition dictionary file on the HDD 7. Record.

Instead of inputting characters from the keyboard, it is also possible to register by directly inputting a character code.

According to this dictionary creation method, not only printed characters but also single handwritten curly characters can be easily registered in the dictionary.

As for the number of characters in one line, the accuracy improves as the number of characters increases, but the processing time for character pattern data increases, so about 10 characters is appropriate.

As explained above, according to the present invention, characters used in an optical character reading device! ! No special device is required to create a common dictionary, and anyone can easily create and maintain a dictionary using the optical character reading device itself. Furthermore, even handwritten curly characters can be registered in the dictionary without any special restrictions.

Then, either a zero-character pattern created from one line of read characters or one line of characters is displayed, making it easier for the operator to recognize the characters.

[Brief explanation of the drawing]

FIG. 1 is a flow diagram showing the procedure of a dictionary creation method according to the present invention. FIG. 2 is an external perspective view of a document processing system that is an embodiment of the present invention. FIG. 3 is a block diagram of the same; FIG. 4 is a flowchart showing the operation related to character reading; FIGS. 5 to 8 are explanatory diagrams accompanying the explanation of the operation in FIG. 4; FIG. 3 is a flowchart showing details of character font discrimination processing in the figure. FIG. 10 is a flowchart showing details of the character code determination process in FIG. 4. FIG. 11 to FIG. 30 are flowcharts for explaining various processes related to creation and maintenance of a dictionary for character recognition. FIGS. 31 to 33 are explanatory diagrams accompanying the explanation of the dictionary registration process shown in FIG. '50. A... Original 1... Keyboard 2... Image scanner 3... CRT display device [4... Printer 5...
・Main unit 6 ・Floppy disk device 7・
...Hard disk device 10...Control unit (CPU) M1 figure figure 5 header section figure 9 figure 31 figure 32 figure 33

Claims (1)

[Claims] 1. An optical character reader that scans a document with a scanner, captures image information including characters as image data, recognizes characters from the image data, and converts them into character code data. At the same time, a document with a mark indicating the height of the characters in that line is scanned by the scanner to capture the image data, and the dot patterns of each character within the height regulated by the mark are overlapped. Combine or average to create one character pattern, and then
A dictionary creation method characterized by displaying any character in a line, giving a character code corresponding to the character pattern, and registering it in a dictionary file for character recognition.